ЦОС_Заочники2013 / Книги по ЦОС / Дополнительная литература / Основы ЦОС (на англ.яз) / MS9

DSP APPLICATIONS

SECTION 9

DSP APPLICATIONS

■High Performance Modems for Plain Old Telephone Service (POTS)

■Remote Access Server (RAS) Modems

■ADSL (Assymetric Digital Subscriber Line)

■Digital Cellular Telephones

■GSM Handset Using SoftFone™ Baseband Processor and Othello™ Radio

■Analog Cellular Basestations

■Digital Cellular Basestations

■Motor Control

■Codecs and DSPs in Voiceband and Audio Applications

■A Sigma-Delta ADC with Programmable Digital Filter

9.a

DSP APPLICATIONS

9.b

DSP APPLICATIONS

SECTION 9

DSP APPLICATIONS

Walt Kester

HIGH PERFORMANCE MODEMS FOR PLAIN OLD

TELEPHONE SERVICE (POTS)

Modems (Modulator/Demodulator) are widely used to transmit and receive digital data using analog modulation over the Plain Old Telephone Service (POTS) network as well as private lines. Although the data to be transmitted is digital, the telephone channel is designed to carry voice signals having a bandwidth of approximately 300 to 3300Hz. The telephone transmission channel suffers from delay distortion, noise, crosstalk, impedance mismatches, near-end and far-end echoes, and other imperfections. While certain levels of these signal degradations are perfectly acceptable for voice communication, they can cause high error rates in digital data transmission. The fundamental purpose of the transmitter portion of the modem is to prepare the digital data for transmission over the analog voice line. The purpose of the receiver portion of the modem is to receive the signal which contains the analog representation of the data , and reconstruct the original digital data at an acceptable error rate. High performance modems make use of digital techniques to perform such functions as modulation, demodulation, error detection and correction, equalization, and echo cancellation.

A block diagram of an ordinary telephone channel (often referred to as plain old telephone service – or POTS) is shown in Figure 9.1. Most voiceband telephone connections involve several connections through the telephone network. The 2-wire twisted pair subscriber line available at most sites is generally converted to a 4-wire signal at the telephone central office: two wires for transmit, and two wires for receive. The signal is converted back to a 2-wire signal at the far-end subscriber line. The 2- to 4-wire interface is implemented with a circuit called a hybrid. The hybrid intentionally inserts impedance mismatches to prevent oscillations on the 4- wire trunk line. The mismatch forces a portion of the transmitted signal to be reflected or echoed back to the transmitter. This echo can corrupt data the transmitter receives from the far-end modem.

Half-duplex modems are capable of passing signals in either direction on a 2-wire line, but not simultaneously. Full-duplex modems operate on a 2-wire line and can transmit and receive data simultaneously. Full-duplex operation requires the ability to separate a receive signal from the reflection (echo) of the transmitted signal. This is accomplished by assigning the signals in the two directions different frequency bands separated by filtering, or by echo cancelling in which a locally synthesized replica of the reflected transmitted signal is subtracted from the composite receive signal.

9.1

DSP APPLICATIONS

ANALOG MODEM USING PLAIN OLD TELEPHONE SERVICE (POTS) ANALOG CHANNEL

FREQUENCY NOISE

SHIFT

FOUR-WIRE TRUNK

Figure 9.1

There are two types of echo in a typical voiceband telephone connection. The first echo is the reflection from the near-end hybrid, and the second echo is from the farend hybrid. In long distance telephone transmissions, the transmitted signal is hetrodyned to and from a carrier frequency. Since local oscillators in the network are not exactly matched, the carrier frequency of the far-end echo may be offset from the frequency of the transmitted carrier signal. In modern applications this shift can affect the degree to which the echo signal can be canceled. It is therefore desirable for the echo canceller to compensate for this frequency offset.

For transmission over the telephone voice network, the digital signal is modulated onto an audio sinewave carrier, producing a modulated tone signal. The frequency of the carrier is chosen to be well within the telephone band. The transmitting modem modulates the audio carrier with the transmit data signal, and the receiving modem demodulates the tone to recover the receive data signal.

The baseband data signal may be used to modulate the amplitude, the frequency, or the phase of the audio carrier, depending on the data rate required. These three types of modulation are known as amplitude shift keying (ASK), frequency shift keying (FSK), and phase shift keying (PSK). In its simplest form the modulated carrier takes on one of two states - that is, one of two amplitudes, one of two frequencies, or one of two phases. The two states represent a logic 0 or a logic 1.

Lowto medium-speed data links usually use FSK up to 1,200 bits/s. Multiphase PSK are used for 2,400 bits/s and 4,800 bits/s links. PSK utilizes bandwidth more efficiently than FSK but is more costly to implement. ASK is least efficient and is

9.2

DSP APPLICATIONS

used only for very low speed links (less than 100 bits/s). For 9,600 bits/s up to 33,600 bits/s a combination of PSK and ASK is used, known as Quadrature Amplitude Modulation (QAM).

The International Telegraph and Telephone Consultative Committee (CCITT in France) has established standards and recommendations for modems which are given in Figure 9.2.

SOME MODEM STANDARDS

*DOWNSTREAM ONLY, UPSTREAM IS V.34

**UPSTREAM AND DOWNSTREAM

Figure 9.2

The goal in designing high performance modems is to achieve the highest data transfer rate possible over the POTS network and avoid the expense of using dedicated conditioned private telephone lines. The V.90 recommendation describes a full-duplex (simultaneous transmission and reception) modem that operates on the POTS network. The V.90 modem communicates downstream from the central office to the subscriber modem at a rate of 56,000 bits/s using Pulse Code Modulation (PCM). Upstream communication from the subscriber to the central office is at the V.34 rate of up to 33,600 bits/s (QAM).

A simplified block diagram for a V.90 analog modem is shown in Figure 9.3. The diagram shows that the bulk of the signal processing is done digitally. Both the transmit and receive portions of the modem subject the digital signals to a number of DSP algorithms which can be efficiently run on modern processors.

9.3

DSP APPLICATIONS

V.90 ANALOG MODEM SIMPLIFIED BLOCK DIAGRAM

MIXED-SIGNAL FUNCTIONS

×

cosωt

Figure 9.3

The TX input serial bit stream is first scrambled and encoded. Scrambling takes the input bit stream and produces a pseudo-random sequence. The purpose of the scrambler is to whiten the spectrum of the transmitted data. Without the scrambler, a long series of identical symbols could cause the receiver to lose carrier lock. Scrambling makes the transmitted spectrum resemble white noise, to utilize the bandwidth of the channel more efficiently, makes carrier recovery and timing synchronization easy, and makes adaptive equalization and echo cancellation possible.

The scrambled bit stream is divided into groups of bits, and the groups of bits are first differentially encoded and then convolutionally encoded.

The symbols are then mapped into the signal space using QAM as defined in the V.34 standard. The signal space mapping produces two coordinates, one for the real part of the QAM modulator and one for the imaginary part. As an example, a diagram of a 16-QAM signal constellation is shown in Figure 9.4. Larger constellations are used in V.90 modems, and the actual size of the constellation is adaptive and determined during the training, or “handshake” interval when the modems synchronize with each other for upstream or downstream signaling.

9.4

DSP APPLICATIONS

QUADRATURE AMPLITUDE MODULATED (QAM) SIGNAL TRANSMITS 4 BITS PER SYMBOL (16-QAM)

Q

1111

4-BITS/SYMBOL

I

0000

I OR Q CHANNEL

SAMPLING CLOCK

t

Figure 9.4

Used prior to modulation, digital pulse shaping filters attenuate frequencies above the Nyquist frequency that are generated in the signal mapping process. These filters are designed to have zero crossings at the appropriate frequencies to cancel intersymbol interference.

QAM is easily implemented in modern DSP processors. The process of modulation requires the access of a sine or cosine value, the access of an input symbol (x or y coordinate) and a multiplication. The parallel architecture of the ADSP-21xx-family permits all three operations to be performed in a single instruction cycle.

The output of the digital modulator drives a DAC. The output of the DAC is passed through an analog lowpass filter and to the 2-wire telephone line for transmission over the POTS network.

The receiver is made up of several functional blocks: the input antialiasing filter and ADC, a demodulator, an adaptive equalizer, a Viterbi decoder, an echo canceller, a differential decoder, and a descrambler. The receiver DSP algorithms are both memory-intensive and computation-intensive. The ADSP-218x-family addresses both needs, providing sufficient program memory RAM (for both code and data) on chip, data memory RAM on chip, and an instruction execution rate of up to 75MIPS.

The antialiasing filter and ADC in the receiver need to have a dynamic range from the largest echo signal to the smallest. The received signal can be as low as -40dBm,

9.5

DSP APPLICATIONS

while the near-end echo can be as high as -6dBm. In order to ensure that the analog front end of the receiver does not contribute any significant impairment to the channel under these conditions, an instantaneous dynamic range of 84dB and an SNR of 72dB is required.

In order to compensate for amplitude and phase distortion in the telephone channel, equalization is required to recover the transmitted data at an acceptably low bit error rate. In order to respond to rapidly changing conditions on the telephone line, adaptive equalization is required for the V.90 modem receiver. An adaptive equalizer can be implemented digitally in an FIR filter whose coefficients are continuously updated based on current line conditions.

Separation between the transmit and receive signal in the V.90 modem is accomplished using echo cancellation. Both near-end and far-end echo must be cancelled in order to yield reliable communication. Echo cancellation is achieved by subtracting an estimate of the echo return signal from the actual received signal. The predicted echo is determined by feeding the transmitted signal into an adaptive filter with a transfer function that approximates the telephone channel. The adaptive filter commonly used in echo cancellers is the FIR filter (chosen for its stability and linear phase response). The taps are determined using the least-mean- square (LMS) algorithm during a training sequence executed prior to full-duplex communications.

The most common technique for decoding the received data is Viterbi decoding. Named after its inventor, the Viterbi algorithm is a general-purpose technique for making an error-corrected decision. Viterbi decoding provides a certain degree of error correction by examining the received bit pattern over time to deduce the value that was the most likely to have been transmitted at a particular time. Viterbi decoding is computation-intensive. A history for each of the possible symbols sent at each symbol interval has to be maintained. At each symbol interval, the length of the path backward in time from each possible received symbol to a symbol sent some time ago is calculated. The symbol that has the shortest path back to the original signal is chosen to be the current decoded symbol. A complete description of Viterbi decoding and its implementation on the ADSP-21xx- family of processors is given in documentation available from Analog Devices (Reference 2).

Figure 9.5 shows a comparison between V.34 and V.90 modems. Note that in the case of V.34 (Figure 9.5A), the communication is between two analog modems. This requires and ADC/DAC in both the transmit and receive path as shown in the diagram. The V.90 system requires an all digital network and a V.90 digital modem as shown in Figure 9.5B. Note that the second ADC/DAC combination is eliminated, thereby allowing the faster downstream data rate of 56Kbits/s. Downstream communication to the V.90 analog modem uses 64Kbits/s PCM data, which is standard for all digital telephone networks. This serial data is converted into a pulse amplitude modulated (PAM) signal (8-bits, 8kSPS) using an 8-bit DAC. The signal from the DAC to the analog modem is therefore a 256K constellation with no imaginary component, i.e., the analog modem receiver must detect which one of the 256 levels is being sent during the symbol interval.

9.6

DSP APPLICATIONS

The V.90 standard allows downstream data rates of up to slightly less than 56Kbits/s, and upstream data rates of up to 33.6Kbits/s (V.34). The V.92 standard will allow 56Kbits/s transmission in both directions.

V.34 VERSUS V.90 MODEMS

33.6K bps

Figure 9.5

REMOTE ACCESS SERVER (RAS) MODEMS

Rapid growth and use of the Internet has created a problem in that there are more users trying to get on the Internet than there is equipment to accommodate all these users. Internet Service Providers (ISPs) like America On Line purchase modem equipment so their customers (referred to as subscribers) can remotely access a network (like the Internet from home). This application of accessing a network from a remote location is called Remote Network Access. The equipment used in this application is called a Remote Access Server (RAS) as shown in Figure 9.6. The Remote Access Server is made up of many modem ports; each modem port can connect to a different user. The RAS can use analog modems, which connects to a POTS line, or digital modems which are compatible with T1, E1, PRI, or BRI lines. Digital modems are used in most RAS systems since they are more efficient for 8 ports or more.

Network access equipment enables individuals, small offices, and traveling employees to connect to corporate networks (Intranets) and the Internet. Internet Service Providers use devices called concentrators to connect their telephone access lines to their networks. These concentrators are also referred to as Remote Access Servers. The rapid growth in the use of the Internet and Intranets has created a tremendous demand for modem equipment.

9.7

DSP APPLICATIONS

INTERNET GATEWAY USING

REMOTE ACCESS SERVER (RAS) MODEM

DATA

ROUTER

Figure 9.6

Remote access servers accommodate employees and small offices/home offices (SOHO) wishing to connect individual computers to LANs (Local Area Networks) or Intranets. If a remote access server is installed in a corporate LAN, remote users can access the network in such a way that their computers appear to be directly connected to the LAN. This allows them to work from a remote location as if they were sitting at their desk in their office.

The ADSP-21mod870 acts as a bridge between the voice-based continuous connection switched network and the data based IP network as shown in Figure 9.7. The high speed DMA interface and large on-chip RAM of the ADSP-21mod870 allows it to be configured to handle a large variety of tasks. The software with the ADSP-21mod870-100 can be configured for modem calls or HDLC (high bit-rate digital subscriber line) processing of digital ISDN (integrated services digital network) based calls. Since the ADSP-21mod870 is an open platform, other functions can be loaded by users. Examples of other functions include voice over internet and FAX over internet. In these applications, the ADSP-21mod870 is a gateway for voice network users to save toll charges by routing their call over the IP network. The ADSP-21mod870 DSP uses the ADSP-218x 16-bit fixed-point core and is code compatible with other members of the ADSP-21xx family.

As the number of remote network users has grown, capacity over the telephone central office switched network has often become strained. These bottlenecks often occur when thousands of calls from a metropolitan area are switched to a single point of presence (POP). To eliminate these bottle necks, RAS equipment can be pushed outward from the POP toward the edges of the switched network as shown in Figure 9.8. When RAS equipment is located in the switching center for local exchanges, data calls can be separated from voice calls, eliminating the strain on

9.8